[CRUNCH-577] NumberFormatException when parsing dfs.block.size

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.8.0, 0.8.1, 0.9.0, 0.8.2, 0.10.0, 0.8.3, 0.8.4, 0.11.0, 0.12.0
• Fix Version/s: 0.14.0
• Component/s: IO
• Labels: patch
• Assignee: Josh Wills
• Reporter: Tomáš Čechal
• Created: 10/Nov/15 10:42
• Updated: 08/May/16 04:14
• Resolved: 11/Nov/15 00:09

Description

When using file size abbreviations (like "128m") for the HDFS configuration property "dfs.block.size" the Crunch job crashes with a NumberFormatException. According to the Hadoop documentation (https://hadoop.apache.org/docs/r2.6.0/hadoop-project-dist/hadoop-hdfs/hdfs-default.xml), this style of abbreviations should be supported.

The problem occurs at line 38 in CrunchCombineFileInputFormat.java when the configuration property is parsed using the getLong() method instead of getLongBytes() method. Furthermore, obsolete configuration key "dfs.block.size" is used instead of "dfs.blocksize" (see https://issues.apache.org/jira/browse/HDFS-631) which leads to a warning message being emitted when starting a MR pipeline.

The proposed solution discussed on the crunch-users mailing list (http://mail-archives.apache.org/mod_mbox/crunch-user/201511.mbox/browser) is to use the getLongBytes() method and the new config key "dfs.blocksize".